Usage
Usage is counted in the model's billing unit and reported on every answered request. Money is never in usage: the cost is x-hips-cost-micros, or on a streamed answer the hips object of the final usage chunk, and the usage record.
usage
{"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0, "prompt_tokens_details": {"cached_tokens": 0}, "completion_tokens_details": {"reasoning_tokens": 0}}}prompt_tokens, completion_tokensCounted by the serving engine's tokenizer for the pinned model. These are the billable units at the model's separate input and output rates.
total_tokensThe sum; not a billing unit.
prompt_tokens_details.cached_tokens0on every answer today: HIPS keeps no prompt cache yet, so every input token is billed at the input rate (Prompt caching).prompt_tokens_details.cache_write_tokensAlways
0: writing a cache costs nothing.completion_tokens_details.reasoning_tokensOn models advertising
reasoning; counted insidecompletion_tokens.
usage
{"usage": {"units": 2, "unit": "classification", "billed_units": 1}}unit names the task's unit. On a classification units counts the inputs and billed_units the inputs that were answered. On an embedding or a rerank the unit is second: units and billed_units are the started seconds of the request's compute time, at least one, and input_tokens the tokens the gateway counted with the model's own tokenizer, a figure that is never the bill (H109, decided 2026-11-05). POST /v1/embeddings keeps OpenAI's usage (prompt_tokens, total_tokens), and POST /v1/rerank says the same in Cohere's meta.billed_units.seconds and meta.tokens.input_tokens. On an image task the unit is image: units counts the images asked and billed_units the images delivered, each billed at the model's price for its size (Image generation and editing); OpenAI's image routes answer no usage, and the cost is x-hips-cost-micros.
| Task | Unit | What is counted | Priced now |
|---|---|---|---|
chat (chat completions, Responses, Messages) | tokens | Input and output tokens at separate rates; cached input at its own rate, though no cached token is billed today | Yes |
classification | classification | One item evaluated once under the task's input conditions is one classification, priced per 1,000 | Yes |
embedding | second | Started seconds of the request's compute time, at least one (H109, decided 2026-11-05). The tokens counted are a figure only, never the bill | Yes |
rerank | second | Started seconds of the request's compute time, at least one (H109); every document is scored, whatever top_n returns | Yes |
detection | detection | Detections | No: not served until priced |
transcription, translation | audio_second | Seconds of input audio | No: not served until priced |
speech | character | Input characters | No: not served until priced |
image_generation, image_edit | image | Output images delivered, each at the model's fixed price for its size; an image the content checks withhold is not billed (Image generation and editing) | Yes |
video_generation | video_second | Seconds of output video | No: not served until priced |
- Task
chat(chat completions, Responses, Messages)- Unit
- tokens
- What is counted
- Input and output tokens at separate rates; cached input at its own rate, though no cached token is billed today
- Priced now
- Yes
- Task
classification- Unit
classification- What is counted
- One item evaluated once under the task's input conditions is one classification, priced per 1,000
- Priced now
- Yes
- Task
embedding- Unit
second- What is counted
- Started seconds of the request's compute time, at least one (H109, decided 2026-11-05). The tokens counted are a figure only, never the bill
- Priced now
- Yes
- Task
rerank- Unit
second- What is counted
- Started seconds of the request's compute time, at least one (H109); every document is scored, whatever
top_nreturns - Priced now
- Yes
- Task
detection- Unit
detection- What is counted
- Detections
- Priced now
- No: not served until priced
- Task
transcription,translation- Unit
audio_second- What is counted
- Seconds of input audio
- Priced now
- No: not served until priced
- Task
speech- Unit
character- What is counted
- Input characters
- Priced now
- No: not served until priced
- Task
image_generation,image_edit- Unit
image- What is counted
- Output images delivered, each at the model's fixed price for its size; an image the content checks withhold is not billed (Image generation and editing)
- Priced now
- Yes
- Task
video_generation- Unit
video_second- What is counted
- Seconds of output video
- Priced now
- No: not served until priced
The embeddings and rerank routes are on Embeddings and rerank. A task type with no price kind is not served until it is priced. Every setting that changes the charge is a named parameter of the task's request, is shown in the Playground beside the measured usage, and is carried in the quote. The unit is set per model in the catalog.
A request sent to a router is billed exactly as the model the router chose, at that model's quote, rates, pool and units, with no router fee (what it costs).
One per request. It holds the request id, timestamps (received, accepted, first byte, completed), the key, the workspace, the pinned id, the deployment if any, task, pool, lane, pricing mode, region, the quote, units in and out, the rate version, the cost, the outcome (completed, failed, canceled, interrupted), the error code if any, the attempt count, latency, the client's address and user agent for abuse control, and the sizes of the body and the output; under a router, the router, its preference and how it chose. It holds no prompt, no input, no output. The Console's Usage and activity view shows and exports it.
Related sheets
- Sheet
- 18 of 20
- Detail
- T
- Updated
- 2026-11-16